I/O rivals declare truce, plan new architecture

The fight brewing over who should set the standard for forthcoming I/O architectures ended before it ever really began, saving users the hassle of deciding between at least two different channel-based I/O designs.

At issue was the way equipment such as disk systems and network cards plug in to servers. Earlier this month, representatives from the opposing Next Generation I/O (NGIO) and Future I/O camps agreed on a compromise, called System I/O, that officials say combines the best of both specifications.

The agreement calls for the development of a minimum standard that will be used as a starting point from which a unified standard can be developed and upon which systems vendors will be able to build. System I/O development will be governed by a new standards group guided by IBM and Intel. Other participants in the System I/O group include Sun Microsystems, Dell Computer, Hewlett-Packard and Microsoft.

According to John Miner, vice-president of the communications product group at Intel, the goal is to have System I/O-based products, from entry-level to data centre-class systems, in production sometime in 2001.

“This train is headed for departure in 2001, which means the architecture and development work needs to get going now in order to be on time,” Miner said.

Fortunately, Miner said, the System I/O group will not be starting from scratch, but will cull the best features from the two-plus years of work the NGIO and Future I/O groups have done.

That head start should let the group deliver the specification this year, which in turn will allow developers to offer System I/O-based products in 2001.

The result will be a channel-based I/O that initially will offer from one to 12 wires — each with a throughput of 2.5GB/sec. — and will be able to span multiple CPU generations, Miner said.

Because the new specification will be channel-based, information can flow back and forth in its own separate channel more quickly than with bus architecture. Channels are also more scalable than buses and less expensive.

What that will bring to future systems, Miner said, is an I/O that breaks through the limitations of the existing PCI standard, allowing the connectivity needed to build modular, scalable systems with such features as hot-swappable I/O.

James Gruener, an analyst at the Aberdeen Group in Boston, said the compromise will benefit both the NGIO and Future I/O backers, and more important, customers.

“It’s a good move for the suppliers and industry at large because of complexity that’s out there trying to get to these new standards,” Gruener said.

However, that complexity will not go away completely with the release of System I/O, because systems vendors will maintain control of their IP licences and a certain degree of independence in building upon the specification.

Another key to the agreement is the migration path being set forth, which will allow customers the freedom to decide when they will upgrade to the new architecture.

According to Gruener, that should be music to customers’ ears.

“It’s probably the best thing for customers because it will be a lot easier for them to do a migration when they’re ready,” Gruener said.

One analyst suggested that the Future I/O group, begun by HP, IBM and Compaq Computer, may have felt it simply could not compete against the Intel-backed NGIO development effort.

“Future I/O people gave up the idea that they all will be equal partners,” said Jonathan Eunice, an analyst at Illuminata Inc. in Nashua, N.H. “Intel will have more say than Compaq or HP. And Intel gave up the idea of [dominating] the standard-setting process.”